OpenAI GPT-5.6 delayed after White House approval request

OpenAI is delaying the public release of its next generation models, GPT-5.6, at the request of the Trump White House, the company confirmed on Friday. The company will first share GPT-5.6 with a small set of customers who are preapproved by the US government, then work with the administration to slowly expand access.

What happened to the GPT-5.6 release?

OpenAI paused a broad launch and will initially provide GPT-5.6 only to a vetted group of customers, expanding access over the coming weeks. The company told customers it hopes to make GPT-5.6 available to everyone in the coming weeks, but for now the models will be shared first with a small set of users preapproved by the US government.

OpenAI said the delay followed a request from the White House and that it will work with the administration to develop the cyber Executive Order framework and a repeatable process for future model releases. Executives said they could not share exact details of the approval mechanics: the company sends the government a list of customers and then gets feedback.

How is the White House influencing access?

The White House asked OpenAI to stagger the release after issuing an executive order setting up a voluntary process for labs to share models with the government 30 days ahead of a broader release. The executive order, signed earlier this month by President Trump, directed a voluntary process and included a carve-out stating the government would not make that process a de facto licensing regime.

OpenAI said no such voluntary framework exists yet, and the company described the current arrangement as an interim period where working with the US government on an AI model launch "doesn’t seem all that voluntary." The White House also recently sent an export control directive to Anthropic, an action that prompted Anthropic to take its most advanced AI models offline for all customers. The request to OpenAI came two weeks after that directive.

What is GPT-5.6 and who gets it first?

GPT-5.6 will be offered in three flavors: Sol, the most capable version; Terra, a middle‑tier version; and Luna, a fast and affordable version. OpenAI says GPT-5.6 Sol is its most capable model yet on benchmarks testing cybersecurity, biology, and agentic abilities. The company also says it has a "layered safeguard stack" intended to prevent malicious uses such as cyberattacks.

OpenAI plans to broaden the set of customers next week to include some international partners, but the initial tranche will be limited to those preapproved by the US government. Executives declined to disclose how customers are selected beyond saying they send a list to the government and receive feedback.

Why it matters

This pause ties a commercial model rollout directly to White House review, shifting how frontier models reach users. The executive order’s 30‑day sharing window and the government’s recent export control pressure on Anthropic together create a new operating environment for US AI labs: access timing and customer lists can now be shaped by national cybersecurity concerns, not only product readiness or market demand.

The move affects developers, enterprises, cyber defenders, and international partners who the company said need access to the tools, and it places OpenAI in the middle of a broader policy push that has moved from deregulation toward active government scrutiny.

What to watch

Watch whether the White House publishes the promised voluntary process for 30‑day model sharing and how that framework defines review criteria. Also track OpenAI’s promised timeline: the company said it will broaden access next week and hopes to make GPT-5.6 generally available in the coming weeks. The pace and scope of that expansion will indicate whether this government review remains temporary or becomes a recurring step for future releases.